ewijaya,
Ok - I had an idea in my head that didn't pan out the way I expected. I would be interested in seeing what I ended up with benchmarked with some real sample data though.

sub remove_poly {
    my ($list, $threshold) = @_;
    return map {
        my $lim = $threshold * length $list->[$_];
        my (%cnt, $excede);
        for ( unpack 'U*', $list->[$_] ) {
            $excede = 1, last if ++$cnt{$_} > $lim;
        }
        $excede ? () : $list->[$_];

    } 0 .. $#$list;
}
[download]